URI-AWARE USER INPUT INTERFACES FOR THE UNOBTRUSIVE ... · URI-AWARE USER INPUT INTERFACES FOR THE...
Transcript of URI-AWARE USER INPUT INTERFACES FOR THE UNOBTRUSIVE ... · URI-AWARE USER INPUT INTERFACES FOR THE...
IADIS International Journal on Computer Science and Information Systems
Vol. 13, No. 2, pp. 62-75 ISSN: 1646-3692
62
URI-AWARE USER INPUT INTERFACES
FOR THE UNOBTRUSIVE REFERENCE
TO LINKED DATA
André Langer, Christoph Göpfert, and Martin Gaedke
Chemnitz University of Technology, Computer Science Department, Chemnitz, Germany.
ABSTRACT
Using appropriate entity URIs is a crucial factor for the success of semantic-enabled applications for data management and data retrieval. Especially data applications that collect data to build knowledge graphs rely on correct concept identifiers in favor of ambiguous literals. This collection involves human interaction in the web frontend without annoying the user. But appropriate user interfaces for this task are still a challenge. In this article, we focus on the design of form elements that unobtrusively allow input data both for human and machine interaction from a semantic point of view. Motivated by web-based scholarly document-submission systems, we first present a brief current-state analysis on the support of
semantic input operations, investigate how these users input interfaces can be improved for concept linking purposes with an auto-suggestion behavior and finally evaluate with a proof-of concept implementation and user survey the advantages and acceptance of our approach.
KEYWORDS
Linked Data, User Interfaces, Usability, Data Management, Web Components
1. INTRODUCTION
Resources in the World Wide Web are traditionally published in a document-centric fashion.
Taking the research community as an example, scientists publish their findings by digitally
uploading a paper that contains natural language floating text via a document submission
system, and may also add additional meta information such as a list of authors and their
affiliation, categorization tags or other publishing information. Other researchers can then find
and access this document information in the result list of a search-based environment by entering
similar keywords. For the last decades, this approach has been easy, sufficient and convenient.
But the amount of information on the World Wide Web increases rapidly and the demand for
URI-AWARE USER INPUT INTERFACES FOR THE UNOBTRUSIVE REFERENCE TO LINKED
DATA
63
new strategies and services requires to better publish, manage and retrieve particular knowledge
in a structured fashion.
The Semantic Web community therefore proposed knowledge graphs as a graph-based
organization of linked information in order to turn the traditional document-centric web into an entity-centric web of data (Shadbolt, Berners-Lee, & Hall, 2006). Multiple application domains
already profited from such an entity-based knowledge representation as content can be found
and filtered on a fine granular entity level, even in different representations.
Though, human interaction is often needed to add or modify information in such a
knowledge graph. Thereby, 1. each entity related to a particular resource has to be described
separately 2. with a persistent identifier that is 3. well-known to represent this particular
real-world concept. Simple literal text input is usually not appropriate to accomplish this task as
different spellings and interpretations could be used. So, human users would have to deal with
the provision of unambiguous persistent identifiers to create business value but which will also
be a very tedious activity.
A good user interface experience (UIX) is needed to lower the barrier and to motivate users to provide correct Linked Data references. Assisting the user during this input activity by
presenting suitable information and hiding complex or unnecessary data in the current context
is one possible approach. In web development, the HyperText Markup Language v5 (HTML5)
therefore already provides a set of simple input elements. These input controls are still used in
current web forms for simple literal text input although they could be enriched to better support
the reference of external Linked Data. The unobtrusive usage together with semantic entity URIs
is hardly considered.
In this article, we extend previous work (Langer, Göpfert, & Gaedke, 2018) dealing with the
research question how to improve the user motivation and experience in Web-enabled
applications with Web Engineering methodologies to let humans provide correct entity URIs
without changing their regular interaction habit. Our contributions are:
1. We provide evidence through a comparison of web forms in established document-oriented
data collection system, that referencing related entities with literal input or local IDs is still
dominating and represents an obstacle for interoperability
2. We coin the term of URI-aware user input interfaces as an extension of traditional HTML
text input elements with an auto-suggestion functionality that present human-readable
content as well as machine-readable persistent identifiers on client and server side
3. We conduct an experimental user survey to show that URI-aware user input interfaces can
considerably increase the quality of provided Linked Data references while maintaining the
traditional interaction user satisfaction
The remainder of the article is structured in the following way. In section 2, we use the scenario of research submission systems as a motivating example, check the current support on
existing web platforms and infer requirements on suitable user interfaces. Section 3 presents
general characteristics of a URI-aware input element and discusses the functional extension of
different types of HTML elements to turn them into URI-aware components. In section 4, we
implement the presented approach as a proof of concept and evaluate in a usability study the
opinion of multiple users. Section 5 contains the Related Work and sets our contribution into
relation to other findings within the Linked Data community. In the last section, we summarize
the entire topic and outline possible future extensions.
IADIS International Journal on Computer Science and Information Systems
64
2. USER INTERFACE SUPPORT FOR REFERENCING
RELATED LINKED DATA ENTITIES
We first provide a common scenario from the academia as a motivating example to illustrate
the problem domain of our research focus, which we will later also use for evaluation purposes.
2.1 A Motivating Use Case
In a traditional research publication scenario, a user Jane has written a scientific publication and
wants to submit this document so that other researchers can find it. So, she uploads a PDF file
and specifies manually further meta data such as the names of contributing authors, the
publication title and abstract, keyword tags, language and license information and others. This
research publication can then be found on web platforms like Bibsonomy, Easychair, Google
Scholar, Mendeley, ResearchGate or Zenodo by other users.
For that purpose, such a document submission system presents the user a web page with an
input form where the concrete document can be uploaded and optionally further processed on
server-side to extract additional meta data in an automated fashion, and additional meta data can
be entered.
Figure 1. Meta data input in a web form and possible resulting, desired knowledge graph structure
This extracted document data together with implicit document meta data and the manually
provided explicit meta data by the user can then be used as the base for search operations
performed by other users the find relevant resources such as the provided document file for
consumption. Traditionally, these search operations are normally text-based as users enter
search terms as strings to find related content that either occurs in the resource itself or in the associated meta data. Normally, this approach returns adequate results, but can lead to
false-positive hits that either contain text with ambiguous meanings (e.g. looking for
“operation” can refer to mathematical operations as well as to documents dealing with
activities) or entities with same strings (e.g. searching for authors with the name Jones). In
addition, it is difficult to retrieve related contributions on a more detailed level than the
publication title or keywords of such a paper (such as the interest in a particular methodology
or result artifact type). It would create value if the publication and its content is described in an
unambiguous, structured fashion to allow better retrieval operations and to improve visibility
and reuse.
URI-AWARE USER INPUT INTERFACES FOR THE UNOBTRUSIVE REFERENCE TO LINKED
DATA
65
An accepted means to accomplish this objective is the construction of knowledge graphs as
depicted in Figure 1. Therefore, persistent Uniform Resource Identifiers (PID/URI) are required
to represent distinct RDF entities. Identification schemes already exist for common items, such
as ArXivID, DOI, PMID or PURL for representing data sets or documents, Dublic Core vocabulary terms for describing bibliographical elements, or FOAF and ORCID for symbolizing
human entities. To simplify naming, we will refer to this entire group of identifiers in the
following as URIs, which also encompasses the term persistent identifiers (PID). These URIs
are representing a comprehensive real-world concept with a unique identifier, but although they
are readable they are hard to remember and abstract for a human user.
Users are accustomed to enter literal text content in an input form instead of providing URIs.
To solve this discrepancy in a semantic-aware web application, users were either explicitly
imposed to enter an explicit identifier, which involved tedious Copy&Paste operations and
commonly resulted in a bad response rate for optional data, or they were asked to simply enter
natural language text, and consecutively a mapping operation was performed in the background
to transform the entered word to a representative URI. But as entity recognition and mapping is not always reliably solving ambiguities or able to find a representation for unknown concepts at
all, the challenge is to both gather explicit Linked Data references via standardized input
interfaces as well as to motivate human users to provide this explicit data. In the following, we
will concentrate on the well-grounded research of this UI improvement.
2.2 An Analysis of Current User Input Interfaces
As a starting point, we first provide an analysis and comparison of current-state web user input
interfaces from existing document-collection systems in the research domain. We are interested,
to which degree established submission platforms already take care of a semantic data input
management and how their user interfaces support this activity. Six service providers for
research data management were conscientiously chosen for this scenario, in detail Bibsonomy,
Easychair, Google Scholar, Mendeley, ResearchGate and Zenodo. Although these systems
differ in their focus area, similarities can be identified when simply concentrating on the
submission form user interface and the provided input elements. As shown in figure 2, these
user interfaces are still dominated by plain text input elements such as input fields and text areas.
They are not only used for string input operations for individual floating text such as a document
title or description, but also for gathering entity-related meta information such as author names and affiliation, keywords to represent categories, or license information. It appears odd that
platform providers attempt to collect additional, structured data for better retrieval operations
from a limited value range but allow an unrestricted input in many cases where different
spellings impede consecutive business operations. It remains unclear, if additional effort for
more sophisticated user interfaces was simply not invested, or if linking of data was not in the
application focus in the past. In contrast to our previous analysis (Langer et al., 2018), the
situation improved in such a way, that commercial providers already enhanced their web form
interfaces during the last four months. The application providers Mendeley and ResearchGate
started to reference entities such as authors and licenses in a more structured fashion, however,
they still use local ID values for their representation or do not provide an ID on frontend side at
all.
IADIS International Journal on Computer Science and Information Systems
66
Bibsonomy Easychair
Google Scholar Mendeley
ResearchGate Zenodo
Figure 2. Meta data user input interfaces in current document submission systems (November 2018)
Table 1 provides a feature matrix to compare multiple supported features among the selected
application platforms with a focus on Linked Data input management in real-world document
submission applications. Relevant assessment criteria are formulated in the first column.
Platforms, that do not support this criterion at all, are indicated with (-), with (o) if they partially
support it, with (+) if a regular plain text input is possible and with (++) if they support it with
a sophisticated UI control.
URI-AWARE USER INPUT INTERFACES FOR THE UNOBTRUSIVE REFERENCE TO LINKED
DATA
67
Table 1. Comparison of document-centric submission web UIs of selected application providers
Bib-
sonomy
Easy-
chair
Scholar
Men-
deley
Research
gate
Ze-
nodo File Submission ? + + O + + + URI/DOI for file? - - - + ++ + Author meta data
input? + + + ++ ++ +
URI for Author? - O - O O + Title input? + + + + + + Abstract /
Descriptive text? - + - + + +
Keyword/Tag/Cate
gory input? + + - + - +
URI for keywords? - - - - - - Additional meta
data? + O + + + +
URI for additional
meta data? - - - - - -
Language input? + - - - - + URI for language? - - - - - - License input? - - - ++ ++ + URI for license? - - - - - - Publication detail
data? + (+) + + - +
URI for
Publication detail
data?
O - - - - O
Explicit document
artifacts? - - - - - -
URI for explicit
document artifacts? - - - - - -
Reference data
input? - - - ++ ++ +
URI for reference
data? - - - + + -
As shown, all user interfaces allow the submission of a file artifact together with common
bibliographical data such as a title, author list, description, and category. For identifiable
real-world entities such as an author, some user interfaces already ask for the selection from a
list of known author names, but without any globally valid URI reference. For other entities
such as categories, this strategy is not similarly followed where only literal input is commonly
expected. Other relevant meta information such as date or language information is often neither
requested at all nor referenced in an easy-processable way. At least, further publication
information regarding the conference, journal or year is often requested in a literal fashion. It
would also make sense to reference the location of the publication in an entity fashion. More
fine-granular structured information for document artifacts such as contained definitions, code
snippets, figures, tables, or other sections could not be found at all. Two applications at least allowed the specification of related contributions.
IADIS International Journal on Computer Science and Information Systems
68
The analysis shows, that concealing the management of Linked Data URIs by instead
presenting readable labels is not the common case in web forms so far for input interfaces.
Although entities such as author names can already be specified as separated entities in favor of
plain text, application providers then handle these objects with local, internal IDs. Other meta data information is still requested via plain text input fields and managed as a literal string.
So, the situation for gathering structured, semantic data in a systematic fashion is not ideal
in order to build knowledge graphs from this data for semantic web applications, where data can
be linked together and set into relation on a more detailed level than simply based on file
resource URIs. To improve this situation, we define the following list of requirements on user
input interface elements in applications dealing with Linked Data:
REQ1: Human-readable literal text can be entered or selected so that the existing interaction
habit of a user is not changed
REQ2: One or multiple corresponding URI value(s) are returned for the provided entity label
to the application
REQ3: Key-Value mappings shall be either statically provided or dynamically retrieved from an appropriate remote web service
REQ4: Optionally, an explicit entity URI shall be providable if none of the presented
mappings represent the entity of interest
REQ5: Existing standard web UI components shall be used
3. URI-AWARE INPUT ELEMENTS FOR IMPROVED
SEMANTIC DATA HANDLING
Based on our research question, how users can be motivated to provide Linked Data references
with existing web input elements that transparently manage entity URI values, we present in
this section a well-grounded approach on how to enrich standard input elements available in W3C’s HTML5 standard with LD URI support capabilities for several conceptual use cases.
3.1 Characteristics of URI-aware Input
In the following, we refer to frontend user input interfaces that are capable of managing
URIs/PIDs as input values for human readable entity labels as URI-aware UI. They shall allow a better semantic data management starting already with the input operation step while also
supporting a human user with assistance functionalities to get correct entities URI for Linked
Data relationships. We thereby assume, that entity URIs shall be favored for RDF object
definitions instead of simple literals, however it does not imply, that URI-aware UI shall be used
for all kind of input data because it makes no sense to represent each individual information
piece with a persistent identifier.
URI-aware user input interfaces have the following characteristics:
UI interaction based on labels instead of URIs
Auto-suggestion functionality with a particular scope of already known data entities
Provision of corresponding URIs as a data value to other frontend components
Fallback option, if none of the known entities matches the desired object
URI-AWARE USER INPUT INTERFACES FOR THE UNOBTRUSIVE REFERENCE TO LINKED
DATA
69
3.2 Enriching Common Input Elements
In this section, we discuss several types of W3Cs HTML5 input elements and the procedures to
turn them into URI-aware user input interfaces.
3.2.1 Input Text
In HTML5, the <input> element natively enables a user to manually input a text string. Such a
string can refer 1. to a plain literal information as well as 2. to a public general URI; in our example scenario e.g. an author name or a topic (keyword). Gathering in case 1 literal content
is an already well-understood task. In order to better process the provided data in a semantic
fashion, web developers are encouraged to consider the explicit specification of corresponding
data attributes (such as datatype or content language). Linking in case 2 to a known entity is
interesting for us, as we want to know the identifier of the entity that is referenced. The trivial
approach is to explicitly require the user in an input field to insert a URI for the desired entity,
selectively decorated with a type=”url” attribute and other DCterms properties.
Yet, manually dealing with an explicit URI may result in a bad response rate and the user
interface experience is cumbersome for a user. Instead, the input process should remain familiar
when dealing with Linked Data and simply enhanced with some assisting functionalities. This
can include a mapping between an entity label and a URI via auto-suggestion, either statically or by using a remote endpoint dynamically. I.e., for person entities, we could use the ORCID
API, or services such as Wictionary or ConcetNet for categorization entities. An HTML5 datalist
that is bound to a text input field can be considered as a solution for enriching a basic text input
element with LD URI management capabilities as shown in figure 3. A datalist allows the
definition of a set of input values together with an alternative label that is considered during text
input.
01 <!DOCTYPE html>
02 <html><body><form>
03 <input name="keyword" list="keywords"/>
04 <datalist id="keywords">
05 <option value="http://www.wikidata.org/entity/Q54837" label="Semantic
Web">
06 <option value="http://www.wikidata.org/entity/Q47146" label="User
Interface">
07 </datalist>
08 <input type="submit"/>
09 </form></body></html>
Figure 3. Enrich a text input element with an auto-suggestion for managing Concept URIs
One drawback of this solution is, that the presentation and behavior of the displayed
suggestion varies between different web browsers. Additionally, the URI value is visibly
inserted into the related input element after selecting the intended entity. This can be desired for
editing and verification purposes, but an advanced, better solution is needed if the user shall
simply interact with literal content. In fact, the user does not even have to interact in any way
when the desired entity is already defined in a distinct way.
In particular, the following five cases have to be distinguished in the auto-suggestion process
how a Linked Data URI can be determined based on the literal input provided by the user that describes a specific entity:
IADIS International Journal on Computer Science and Information Systems
70
1. Single bijective match between the entered entity name and the corresponding Linked
Data URI is a correct representation of this object: then, the suggestion does not even
have to be displayed, as the corresponding URI can be retrieved in a distinct way
2. Single bijective match between the entered entity name and a corresponding Linked Data URI but which represents a different concept: then, the user should get a feedback
of the retrieved and likely URI result with a possibility to change the identifier
3. Multiple URI matches for the entered entity name that represent different concepts:
then, the user should get the possibility to select the desired concept
4. Multiple URI matches for the entered entity name that represent the same concept:
then, the user should get the possibility to select the favored URI
5. No matching URI for the entered entity name found at all: then, we need a fallback as
we need a URI in every case to build a knowledge graph:
Ask the user manually for an explicit Linked Data URI
Auto-generate a URI
Map the input data to a pre-defined URI indicating that no corresponding entity URI is known so far
We applied an advanced dropdownlist component to a text input field that can deal with
these cases. Figure 4 shows, that the underlying entity URI does not even have to be visible for
the user in order to select the correct concept that is meant with the entered input string.
Figure 4. Enrich a text input element with a data list auto-suggestion for managing Concept URIs
3.2.2 Selection Elements
Controls like select lists, radio button groups and checkbox groups are other input elements that
allow the selection of predefined options which a user can choose. They can be used to specify
additional meta data such as a document license or language. Values for appropriate Linked
Data URIs are e.g. http://creativecommons.org/licenses/by-* for a license, or
http://lexvo.org/id/* for language information (in favor of IETF language tags).
URI-AWARE USER INPUT INTERFACES FOR THE UNOBTRUSIVE REFERENCE TO LINKED
DATA
71
Figure 5. Three types of URI-aware HTML5 selection elements, exemplarily with one option
In HTML5, these tags allow the built-in definition of key/value pairs. In traditional web
applications, local IDs are commonly used as a value for a key label. Thus, in order to make these UI input elements URI-aware, developers should instead provide carefully selected Linked
Data URIs as a value. They can either be defined statically or retrieved dynamically from a
remote endpoint in the Semantic Web. Providing an underlying global persistent identifier
already on the frontend side increases interoperability and enhances the content processing by
external tools. Additionally, an “Other” option with an explicit URI input field or a “None of
the above” option should be provided as already discussed in section 3.2.1 as best practice.
3.2.3 Referencing Entity Parts
There are entities apart from already discussed concepts that can be used as descriptive meta
data in our document submission scenario. If we want to set data into relation to other related contributions on a fine granular level, this might also involve the reference to a particular part
of the current document such as included figures, tables, definitions, code snippets as well as
bibliographical reference data that is related to the current publication. If the submission is
already identified by a persistent identifier, e.g. DOI, a URI for a particular entity within this
document can easily be created e.g. by using a fragment URI built from the document URI and
a fragment identifier. This applies both to a detailed internal description of the current document
content as well as a reference to external sources. In the first case, a URI-aware user input
interface can rely on preprocessed structural information of the current object, in our case a
document, and provide meaningful labels in suggestions for all identified parts of this object. In
the second case, APIs of publication collection platforms can be consumed to refer to referenced
literature with correct title/URI key/value pairs in favor of simple literal text for referenced
literature.
3.3 Component-oriented Realization
To implement the described characteristics of URI-aware user input interfaces, a
straight-forward reusable approach to extend existing HTML5 interfaces is desirable. W3C
WebComponents allow the component-based usage of URI-aware user input interfaces. Such a component can be defined as shown in figure 6. Uriaware-input is an exemplary realization of
a text field that allows the multiple input of data labels (here representing keywords). When
entering a string, an auto-suggestion overlay is interactively opened. Elements for this
uriaware-list can either be defined statically (static-option) or retrieved from a remote endpoint
(dynamic-option). A combination of multiple options is also possible, If none of the presented
options matches, an alternative-option can be defined and selected, either by specifying a fixed
URI or by providing an explicit URI by the user.
IADIS International Journal on Computer Science and Information Systems
72
01 <!DOCTYPE html>
02 <html><head><script src="uri-aware-wc.js"></script></head>
03 <body>
04 <form>
05 <label for=”keywords”>Keywords</label>
06 <uriaware-input name=”keywords” id=”keywords” multiple-selection=”true”
order=”ascending” alternative-type=”fixed” >
07 <static-option value=”http://dbpedia.org/resource/Linked_Data” label=”Linked Data” />
08 <dynamic-option src="https://www.wikidata.org/w/api.php?action=wbsearchentities&search=" />
09 <alternative-option value="http://dbpedia.org/resource/other" label="None of the above" />
10 </uriaware-input>
11 <input type="submit"/>
12 </form></body></html>
Figure 6. Implementing a URI-aware text input as a WebComponent
4. EVALUATION
In order to evaluate our approach, we first created a prototypical implementation as a
proof-of-concept for URI-aware input element. Based on that, we conducted on a website an
unsupervised experimental survey. With this study, we tested if the provision rate for entity
URIs increased when using improved user interface elements for Linked Data reference input
among multiple participants. The prototypical implementation and the user survey can be
accessed via https://purl.org/net/vsr/semanticui.
We applied the already discussed use case of submitting a scientific document via an input
form on a web page. Therefore, we asked participants to finish six different tasks through
interaction with the provided web user input interface. In the first three tasks, they had to enter
author information together with an appropriate ORCID persistent identifier. In the other three tasks, we asked to provide categories to classify the document content by providing appropriate
keywords and a corresponding entity URI. Each task presented a different user interface to
accomplish this task. First, a traditional literal input without any support, then a plain entity URI
input element, and finally a traditionally looking form enriched with our URI-aware
auto-suggestion approach. After each experiment, all participants were required to answer a set
of provided statements to subjectively assess their personal user experience based on a
five-level Likert rating scale.
Based on (Nielsen & Landauer, 1993), we focused on a small test user group in favor of a
larger set of questions on the interface usability to ask each candidate afterwards. Eight users
participated in total. According to Nielsen et. al, this test set is appropriate for identifying
usability issues and opinions, as results would not differ heavily with a larger participant group.
Statistical distribution: 62% male, 38% female, Age range: 50% between 20-29, 25% between 30-39, 25% between 40-49, Qualification: 50% graduate student, 50% researchers, Expert level:
50% Advanced knowledge, 25% intermediate knowledge, 25% novice.
All of the participants stated to be basically familiar with tradition web form user interfaces
and able to provide appropriate literal meta data such as a prename, surname, organization and
email correctly. One user filled out the country data not correctly. Interestingly, the 88% of all
participants simply ignored the optional ORCID input field where it was possible to provide a
reference for a persistent author identifier, only 1 out of 8 users was able to provide a correct
value for it.
URI-AWARE USER INPUT INTERFACES FOR THE UNOBTRUSIVE REFERENCE TO LINKED
DATA
73
The subjective rating of the statement if they are confident what to enter as a correct value
in such an URI input field also confirmed this. When requesting them to explicitly provide a
Linked Data URI for an author ahead in a second form, all participants were able to perform the
Retrieve and Paste operation (one participant only provided the ID and not a URI). The action step itself was assessed as quite easy, however, 50% argued and agreed that it does not feel
natural to enter URIs for specifying person data. After improving the web form with a
URI-aware user input behavior for an author name, where the ORCID was fetched and exposed
semi-automatically, the provision rate for a provided correct author entity URI increased to 88%,
which is a substantial improvement in comparison to the traditional web form input with a
response rate of only 12%. Interestingly, users nevertheless assessed it only neutral that it would
be more convenient to not getting asked for a URI input.
Similar results were obtained in the second, keyword-oriented test scenario. On average, a
participant provided four keywords in the traditional, literal based input form. Spelling, word
combinations and acronyms varied widely. When entering explicit category URIs in the second
test, 38% were not able or willing to provide this information in a correct way at all, the others participants on average only provided two entity URIs from Wikidata or DBpedia. 72%
disagreed that it is easy and familiar to provide explicit URIs for keywords, instead they find it
more convenient to work with literal text inputs. Then in the third test, the web form with implicit
URI-retrieval in the background showed again that participants provided the same amount of
tags as in the traditional case, this time with corresponding correct URIs, although the subjective
benefit of the suggestion was only rated as neutral on average.
Traditional input form with literal input
Explicitly asking the user to provide Linked Data URIs
Improved regular input form with URI-aware input element
Figure 7. Result overview for a set of related questions from our conducted user survey
0
5
1 2 3 4 5 6
It is usually easy for me to fill out this kind of input
forms on websites
0
5
1 2 3 4 5
I was self-confident what to enter as a correct value in the ORCID form input field
0
5
1 2 3 4 5
It was easy for me to fill out this keyword input element
05
1 2 3 4 5
In my opinion, the form input interface was strange to enter author information
0
5
1 2 3 4 5
It was easy for me to enter the correct ORCID URI
0
5
1 2 3 4 5
It is more convenient for me to provide keywords instead of
URIs
0
5
1 2 3 4 5
I wondered how this form differs from the first web
form
0
5
1 2 3 4 5
Not getting asked for an author's ORCID URI was
convenient for me
0
5
1 2 3 4 5
The suggestions helped me to select the correct concept
IADIS International Journal on Computer Science and Information Systems
74
5. RELATED WORK
In the past, the management of Linked Data primarily concentrated on backend aspects by building knowledge graphs and storing semantic data in a graph database. The information for these knowledge graphs was assumed to be either published in an explicit way (Heath & Bizer, 2011), derived from existing text data through Entity recognition (Usbeck et al., 2014), or extracted with NLP or AI technologies (Rizzo, Troncy, Hellmann, & Bruemmer, 2012). Frontend aspects focused on the visualization of Linked Data as knowledge graph (Valentina Ivanova, Patrick Lambrix, Steffen Lohmann, & Catia Pesquita, 2017) or dealt with ways to embed structured data in HTML (Microformats or W3Cs’ RDFa) in a uniform, extractable way (Bizer et al., 2013) by providing well-known attributes in HTML tags. In HTML5, dedicated attributes as well as generic data-* attributes were standardized (Bostock, Ogievetsky, & Heer, 2011). Input interfaces can obviously also be annotated with these attributes (“hcard-input-examples Microformats Wiki,” 2017), yet, the collection of Linked Data references involves also other frontend considerations. Although frontend frameworks like SemanticUI exist, their understanding of semantics focusses primarily on self-explaining element descriptors as class names and less on standardized frontend elements that allow the input and reference of Linked Data entities. Projects like LD-R: Linked Data reactor (Khalili & Graaf, 2017) makes a step in the direction to standardized frontend Flux components to view and edit Linked Data, but favor the presentation of Linked Data instead describing components for input operation. We are convinced that also input interfaces have to be improved, because not all relevant data can automatically be extracted from existing data sources and requires human interaction, at least in a data curation step. Our contribution and LD-R can therefore benefit from each other. Using auto-suggestion for keyword mappings was already suggested by other authors ((Latif, Afzal, Hoefler, Saeed, & Tochtermann, 2009) in a case study, but their approach had a narrower context by only focusing on keywords and concentrated on the construction of SPARQL queries in the backend. Instead, our presented follows a human-centered approach and focuses on assisting the user as the appropriate actor to explicitly enter and distinguish the information that was actually meant to build knowledge graphs with correct data.
6. CONCLUSION AND FUTURE WORK
In this article, we presented URI-aware user input interfaces, that allow the management of data labels together with corresponding Linked Data entity URIs as persistent identifiers in the frontend of semantic-aware web applications in a transparent and inconspicuous way. After discussing characteristics of URI-aware user input interfaces, we described the realization as standardized WebComponents. We implemented a proof-of-concept for URI-aware user input interfaces and used it in a document submission scenario in combination with a user study. The evaluation confirmed, that the provision rate for Linked Data URIs can increase dramatically if users do not have to deal with explicit URIs/PIDs on their own in cases where it is not entirely necessary, even if they are unaware that they currently interact with Linked Data. Our contribution has a relevant impact on the development of new data-driven web applications, as Frontend-related input aspects for Linked Data have not sufficiently been discussed before. We showed that humans are unwilling to working with persistent identifiers if they do not have to, and accurately crafted assisting user interfaces can improve this situation.
In the future, the user experience for our presented solution can still be enhanced. Users reported that typeahead functionalities in combination with contextual constraints on the
URI-AWARE USER INPUT INTERFACES FOR THE UNOBTRUSIVE REFERENCE TO LINKED
DATA
75
suggested data can further improve the UIX by assuring an uninterrupted input operation. Another challenge is the application of these findings to alternative data input interfaces such as voice or gesture on mobile devices.
ACKNOWLEDGEMENT
This work was supported by the grant from the German Federal Ministry of Education and Research (BMBF) for the LEDS Project under grant agreement No 03WKCG11D.
REFERENCES
Bizer, C., Eckert, K., Meusel, R., Mühleisen, H., Schuhmacher, M., & Völker, J. (2013). Deployment of RDFa, microdata, and microformats on the web - A quantitative analysis. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) (Vol. 8219 LNCS, pp. 17–32). Springer, Berlin, Heidelberg. https://doi.org/10.1007/978-3-642-41338-4_2
Bostock, M., Ogievetsky, V., & Heer, J. (2011). D3 Data-Driven Documents. IEEE Transactions on Visualization and Computer Graphics, 17(12), 2301–2309. https://doi.org/10.1109/TVCG.2011.185
hcard-input-examples · Microformats Wiki. (2017). Retrieved July 25, 2018, from http://microformats.org/wiki/hcard-auto-fill-examples
Heath, T., & Bizer, C. (2011). Linked Data: Evolving the Web into a Global Data Space. Synthesis Lectures on the Semantic Web: Theory and Technology, 1(1), 1–136. https://doi.org/10.2200/S00334ED1V01Y201102WBE001
Khalili, A., & Graaf, K. A. de. (2017). Linked Data Reactor: Towards Data-aware User Interfaces. In Semantics2017 Proceedings. Amsterdam. Retrieved from http://research.ld-r.org/papers/Semantics_2017_dataAwareUIs.pdf
Langer, A., Göpfert, C., & Gaedke, M. (2018). F.I.E.L.D.S. – Analyzing form input interfaces for explicit Linked Data handling in document submission systems. In Proceedings of 17th International Conference WWW/Internet (ICWI2018). Budapest. Retrieved from http://www.iadisportal.org/digital-library/fields-analyzing-form-input-interfaces-for-explicit-linked-data-handling-in-document-submission-systems
Latif, A., Afzal, M. T., Hoefler, P., Saeed, A. U., & Tochtermann, K. (2009). Turning keywords into URIs: simplified user interfaces for exploring linked data. ACM International Conference Proceeding Series; Vol. 403, 76–81. https://doi.org/10.1145/1655925.1655939
Nielsen, J., & Landauer, T. K. (1993). A mathematical model of the finding of usability problems. In Proceedings of the SIGCHI conference on Human factors in computing systems - CHI ’93 (pp. 206–213). New York, New York, USA: ACM Press. https://doi.org/10.1145/169059.169166
Rizzo, G., Troncy, R., Hellmann, S., & Bruemmer, M. (2012). NERD meets NIF: Lifting NLP extraction results to the linked data cloud. In CEUR Workshop Proceedings (Vol. 937). Retrieved from http://nerd.eurecom.fr/ui/paper/Rizzo_Troncy_Hellmann_Bruemmer-ldow2012.pdf
Shadbolt, N., Berners-Lee, T., & Hall, W. (2006). The Semantic Web - The Semantic Web Revisited. IEEE Intelligent Systems, 21(3), 96. https://doi.org/10.1109/MIS.2006.62
Usbeck, R., Ngonga Ngomo, A.-C., Röder, M., Gerber, D., Coelho, S. A., Auer, S., & Both, A. (2014). AGDISTIS - Graph-Based Disambiguation of Named Entities Using Linked Data (pp. 457–471). Springer, Cham. https://doi.org/10.1007/978-3-319-11964-9_29
Valentina Ivanova, Patrick Lambrix, Steffen Lohmann, & Catia Pesquita. (2017). CEUR-WS.org/Vol-1947 - Visualization and Interaction for Ontologies and Linked Data (VOILA! 2017). In VOILA. Vienna: CEUR. Retrieved from http://ceur-ws.org/Vol-1947/